llama_cpp 0.15.1 → 0.15.3

data/vendor/tmp/llama.cpp/ggml.h

@@ -468,7 +468,6 @@ extern "C" {
         GGML_OP_SOFT_MAX_BACK,
         GGML_OP_ROPE,
         GGML_OP_ROPE_BACK,
-        GGML_OP_ALIBI,
         GGML_OP_CLAMP,
         GGML_OP_CONV_TRANSPOSE_1D,
         GGML_OP_IM2COL,
@@ -482,9 +481,7 @@ extern "C" {
         GGML_OP_ARGSORT,
         GGML_OP_LEAKY_RELU,
 
-        GGML_OP_FLASH_ATTN,
         GGML_OP_FLASH_ATTN_EXT,
-        GGML_OP_FLASH_FF,
         GGML_OP_FLASH_ATTN_BACK,
         GGML_OP_SSM_CONV,
         GGML_OP_SSM_SCAN,
@@ -520,6 +517,7 @@ extern "C" {
         GGML_UNARY_OP_TANH,
         GGML_UNARY_OP_ELU,
         GGML_UNARY_OP_RELU,
+        GGML_UNARY_OP_SIGMOID,
         GGML_UNARY_OP_GELU,
         GGML_UNARY_OP_GELU_QUICK,
         GGML_UNARY_OP_SILU,
@@ -565,7 +563,8 @@ extern "C" {
     // n-dimensional tensor
     struct ggml_tensor {
         enum ggml_type         type;
-        enum ggml_backend_type backend;
+
+        GGML_DEPRECATED(enum ggml_backend_type backend, "use the buffer type to find the storage location of the tensor");
 
         struct ggml_backend_buffer * buffer;
 
@@ -766,7 +765,8 @@ extern "C" {
     GGML_API           bool ggml_is_3d        (const struct ggml_tensor * tensor);
     GGML_API           int  ggml_n_dims       (const struct ggml_tensor * tensor); // returns 1 for scalars
 
-    GGML_API bool ggml_are_same_shape(const struct ggml_tensor * t0, const struct ggml_tensor * t1);
+    GGML_API bool ggml_are_same_shape (const struct ggml_tensor * t0, const struct ggml_tensor * t1);
+    GGML_API bool ggml_are_same_stride(const struct ggml_tensor * t0, const struct ggml_tensor * t1);
 
     // use this to compute the memory overhead of a tensor
     GGML_API size_t ggml_tensor_overhead(void);
@@ -1074,6 +1074,14 @@ extern "C" {
             struct ggml_context * ctx,
             struct ggml_tensor  * a);
 
+    GGML_API struct ggml_tensor * ggml_sigmoid(
+            struct ggml_context * ctx,
+            struct ggml_tensor  * a);
+
+    GGML_API struct ggml_tensor * ggml_sigmoid_inplace(
+            struct ggml_context * ctx,
+            struct ggml_tensor  * a);
+
     GGML_API struct ggml_tensor * ggml_gelu(
             struct ggml_context * ctx,
             struct ggml_tensor  * a);
@@ -1428,15 +1436,13 @@ extern "C" {
             struct ggml_context * ctx,
             struct ggml_tensor  * a);
 
-    // fused soft_max(a*scale + mask + pos[i]*(ALiBi slope))
+    // fused soft_max(a*scale + mask*(ALiBi slope))
     // mask is optional
-    // pos is required when max_bias > 0.0f
     // max_bias = 0.0f for no ALiBi
     GGML_API struct ggml_tensor * ggml_soft_max_ext(
             struct ggml_context * ctx,
             struct ggml_tensor  * a,
             struct ggml_tensor  * mask,
-            struct ggml_tensor  * pos,
             float                 scale,
             float                 max_bias);
 
@@ -1452,11 +1458,12 @@ extern "C" {
             struct ggml_tensor  * b);
 
     // rotary position embedding
-    // if mode & 1 == 1, skip n_past elements (DEPRECATED)
+    // if mode & 1 == 1, skip n_past elements (NOT SUPPORTED)
     // if mode & 2 == 1, GPT-NeoX style
     // if mode & 4 == 1, ChatGLM style
     //
     // b is an int32 vector with size a->ne[2], it contains the positions
+    // c is freq factors (e.g. phi3-128k), (optional)
     GGML_API struct ggml_tensor * ggml_rope(
             struct ggml_context * ctx,
             struct ggml_tensor  * a,
@@ -1475,10 +1482,11 @@ extern "C" {
             int                   n_ctx);
 
     // custom RoPE
-    GGML_API struct ggml_tensor * ggml_rope_custom(
+    GGML_API struct ggml_tensor * ggml_rope_ext(
             struct ggml_context * ctx,
             struct ggml_tensor  * a,
             struct ggml_tensor  * b,
+            struct ggml_tensor  * c,
             int                   n_dims,
             int                   mode,
             int                   n_ctx,
@@ -1491,10 +1499,11 @@ extern "C" {
             float                 beta_slow);
 
     // in-place, returns view(a)
-    GGML_API struct ggml_tensor * ggml_rope_custom_inplace(
+    GGML_API struct ggml_tensor * ggml_rope_ext_inplace(
             struct ggml_context * ctx,
             struct ggml_tensor  * a,
             struct ggml_tensor  * b,
+            struct ggml_tensor  * c,
             int                   n_dims,
             int                   mode,
             int                   n_ctx,
@@ -1506,18 +1515,41 @@ extern "C" {
             float                 beta_fast,
             float                 beta_slow);
 
-    // compute correction dims for YaRN RoPE scaling
-    GGML_CALL void ggml_rope_yarn_corr_dims(
-        int n_dims, int n_orig_ctx, float freq_base, float beta_fast, float beta_slow, float dims[2]);
+    GGML_DEPRECATED(GGML_API struct ggml_tensor * ggml_rope_custom(
+            struct ggml_context * ctx,
+            struct ggml_tensor  * a,
+            struct ggml_tensor  * b,
+            int                   n_dims,
+            int                   mode,
+            int                   n_ctx,
+            int                   n_orig_ctx,
+            float                 freq_base,
+            float                 freq_scale,
+            float                 ext_factor,
+            float                 attn_factor,
+            float                 beta_fast,
+            float                 beta_slow),
+        "use ggml_rope_ext instead");
 
-    // xPos RoPE, in-place, returns view(a)
-    GGML_API struct ggml_tensor * ggml_rope_xpos_inplace(
+    GGML_DEPRECATED(GGML_API struct ggml_tensor * ggml_rope_custom_inplace(
             struct ggml_context * ctx,
             struct ggml_tensor  * a,
             struct ggml_tensor  * b,
             int                   n_dims,
-            float                 base,
-            bool                  down);
+            int                   mode,
+            int                   n_ctx,
+            int                   n_orig_ctx,
+            float                 freq_base,
+            float                 freq_scale,
+            float                 ext_factor,
+            float                 attn_factor,
+            float                 beta_fast,
+            float                 beta_slow),
+        "use ggml_rope_ext_inplace instead");
+
+    // compute correction dims for YaRN RoPE scaling
+    GGML_CALL void ggml_rope_yarn_corr_dims(
+        int n_dims, int n_orig_ctx, float freq_base, float beta_fast, float beta_slow, float dims[2]);
 
     // rotary position embedding backward, i.e compute dx from dy
     // a - dy
@@ -1525,6 +1557,7 @@ extern "C" {
             struct ggml_context * ctx,
             struct ggml_tensor  * a,
             struct ggml_tensor  * b,
+            struct ggml_tensor  * c,
             int                   n_dims,
             int                   mode,
             int                   n_ctx,
@@ -1538,16 +1571,6 @@ extern "C" {
             float                 xpos_base,
             bool                  xpos_down);
 
-    // alibi position embedding
-    // in-place, returns view(a)
-    GGML_DEPRECATED(GGML_API struct ggml_tensor * ggml_alibi(
-            struct ggml_context * ctx,
-            struct ggml_tensor  * a,
-            int                   n_past,
-            int                   n_head,
-            float                 bias_max),
-        "use ggml_soft_max_ext instead (will be removed in Mar 2024)");
-
     // clamp
     // in-place, returns view(a)
     GGML_API struct ggml_tensor * ggml_clamp(
@@ -1677,12 +1700,24 @@ extern "C" {
             float                 p1);
 
     // nearest interpolate
+    // multiplies ne0 and ne1 by scale factor
     // used in stable-diffusion
     GGML_API struct ggml_tensor * ggml_upscale(
             struct ggml_context * ctx,
             struct ggml_tensor  * a,
             int                   scale_factor);
 
+    // nearest interpolate
+    // nearest interpolate to specified dimensions
+    // used in tortoise.cpp
+    GGML_API struct ggml_tensor * ggml_upscale_ext(
+            struct ggml_context * ctx,
+            struct ggml_tensor  * a,
+            int                   ne0,
+            int                   ne1,
+            int                   ne2,
+            int                   ne3);
+
     // pad each dimension with zeros: [x, ..., x] -> [x, ..., x, 0, ..., 0]
     GGML_API struct ggml_tensor * ggml_pad(
             struct ggml_context * ctx,
@@ -1724,13 +1759,6 @@ extern "C" {
             struct ggml_tensor  * a,
             int                   k);
 
-    GGML_API struct ggml_tensor * ggml_flash_attn(
-            struct ggml_context * ctx,
-            struct ggml_tensor  * q,
-            struct ggml_tensor  * k,
-            struct ggml_tensor  * v,
-            bool                  masked);
-
 #define GGML_KQ_MASK_PAD 32
 
     // q:    [n_embd, n_batch,     n_head,    1]
@@ -1744,12 +1772,14 @@ extern "C" {
             struct ggml_tensor  * k,
             struct ggml_tensor  * v,
             struct ggml_tensor  * mask,
-            float                 scale);
+            float                 scale,
+            float                 max_bias);
 
     GGML_API void ggml_flash_attn_ext_set_prec(
             struct ggml_tensor * a,
             enum ggml_prec       prec);
 
+    // TODO: needs to be adapted to ggml_flash_attn_ext
     GGML_API struct ggml_tensor * ggml_flash_attn_back(
            struct ggml_context * ctx,
            struct ggml_tensor  * q,
@@ -1758,14 +1788,6 @@ extern "C" {
            struct ggml_tensor  * d,
            bool                  masked);
 
-    GGML_API struct ggml_tensor * ggml_flash_ff(
-            struct ggml_context * ctx,
-            struct ggml_tensor  * a,
-            struct ggml_tensor  * b0,
-            struct ggml_tensor  * b1,
-            struct ggml_tensor  * c0,
-            struct ggml_tensor  * c1);
-
     GGML_API struct ggml_tensor * ggml_ssm_conv(
             struct ggml_context * ctx,
             struct ggml_tensor  * s,
@@ -2379,6 +2401,7 @@ extern "C" {
     GGML_API int ggml_cpu_has_avx512     (void);
     GGML_API int ggml_cpu_has_avx512_vbmi(void);
     GGML_API int ggml_cpu_has_avx512_vnni(void);
+    GGML_API int ggml_cpu_has_avx512_bf16(void);
     GGML_API int ggml_cpu_has_fma        (void);
     GGML_API int ggml_cpu_has_neon       (void);
     GGML_API int ggml_cpu_has_arm_fma    (void);